Twenty years ago, Borders Books had 1,200 stores, 35,000 employees, and a reasonable argument that people would always want to browse shelves in person. They were right about people. They were catastrophically wrong about the economics. The internet didn't destroy the book industry — it destroyed that book industry. The people who understood the difference retired early. The ones who didn't lost their jobs.
The same structural shift is happening right now in professional services. Not as a prediction. As a measurable, ongoing fact.
The Numbers Goldman Sachs Doesn't Want to Summarize for You
In March 2023, Goldman Sachs published research estimating that generative AI has the potential to automate tasks equivalent to 300 million full-time positions globally. That number went everywhere. The second half of the same report got far less coverage: the bank's economists projected that AI could increase global GDP by 7 percent and create entirely new job categories at a scale comparable to the commercialization of the internet.
The McKinsey Global Institute's June 2023 analysis found that 60 to 70 percent of current work activities — not jobs, activities — could be automated using technology available today. The same report found that demand for roles requiring advanced technological skills will grow 25 percent by 2030, while demand for physical and manual work in predictable environments continues to decline.
jobs exposed to AI automation
(Goldman Sachs, 2023)
with AI available today
(McKinsey, 2023)
from AI integration
(Goldman Sachs, 2023)
The pattern isn't surprising to anyone who studies labor economics. MIT economist Daron Acemoglu and colleagues have documented it across every major technology wave of the last 150 years: the loom, the steam engine, the assembly line, the ATM, the internet. Each wave displaced specific tasks. Each wave created net new employment in forms that didn't exist before the displacement. The transition window — typically 7 to 15 years — is where the pain concentrates.
We are 2 to 3 years into that window.
Which Roles Are Most Exposed
The World Economic Forum's 2025 Future of Jobs Report, drawing on data from more than 1,000 employers across 55 economies, identifies the occupations facing the steepest structural pressure over the next five years: data entry clerks, administrative assistants, payroll and bookkeeping professionals, legal and financial document reviewers, junior market research analysts, and tier-one customer service functions.
What these roles share isn't low skill. Most require real expertise. What they share is high task predictability with large available training data. A first-year associate doing document review has a well-defined task, abundant historical examples, and no contextual judgment the client is actually paying for. That's not a commentary on the associate's intelligence. It's a description of the task's AI exposure profile.
Roles where human judgment operates in high-ambiguity, relationship-dependent, or real-time physical environments face a different trajectory. The WEF projects net growth in: AI and machine learning specialists, data analysts and scientists, sustainability and ESG professionals, business development executives, and operations managers responsible for implementing new technology workflows.
"The senior role in every field gets more powerful. The entry-level commodity role becomes a software subscription."
The pattern holds across sectors. Legal: document review contracts, complex strategy expands. Finance: data aggregation and modeling compress, senior advisory and cross-domain synthesis grows. Medicine: administrative and coding functions automate, diagnostic nuance and patient relationship deepens. The senior role in every field gets more powerful. The entry-level commodity role becomes a software subscription.
The Historical Precedent Is Unambiguous
ATMs deployed at scale in the United States starting in the 1970s. Conventional wisdom held they would eliminate bank tellers. BLS data tells a different story: the number of bank tellers in America increased through the 1980s and 1990s, peaking near 600,000 in the mid-2000s. Why? ATMs reduced the per-branch cost of a teller, which meant banks could profitably open more branches, which required more tellers.
The internet is estimated to have eliminated 3.7 million U.S. retail positions between 1999 and 2019. The same period created more than five million jobs in logistics, e-commerce operations, software, digital marketing, and platform management — the majority paying more than the retail positions they replaced.
This is not naïve optimism. The transition was genuinely brutal for the people caught in the displaced category. Retail workers in their 50s in 2005 did not smoothly pivot to cloud architecture jobs. The transition costs are real, they fall unevenly, and they deserve honest accounting. The net outcome — across the full labor market — has nonetheless been positive in every comparable disruption on record.
AI will follow the same pattern. The people who assume that precedent applies to others and not to their own profession are making the exact mistake Borders made in 2003.
What the Smartest Executives Are Doing Right Now
The executives pulling away from their peers are not waiting for AI strategy memos. They are running concrete pilot workflows in the highest-volume, most-predictable parts of their operations. They are measuring output per dollar, not headcount. And they are identifying which functions — when augmented by AI — become ten times more valuable rather than replaceable.
Accenture's 2024 research on AI adoption found that companies in the top quintile of AI integration had revenue growth 1.7 times higher than sector averages. The differentiating variable wasn't which tools they deployed. It was the speed of integration into existing workflows and the degree to which human roles were redesigned around AI output — not merely alongside it.
Three moves define the leaders:
Redesign the role, not just the tool.
The firms winning aren't layering AI onto existing job descriptions. They're asking: if AI handles information retrieval and initial synthesis, what should the human be doing with the hour that frees up? In law, it's client strategy. In finance, it's scenario planning. In medicine, it's diagnostic nuance. The answer, in every field, is higher-order judgment.
Promote the people who learn first.
Harvard Business Review's analysis of knowledge worker productivity found that AI-augmented professionals completed complex tasks 25 to 40 percent faster — and the highest productivity gains went to less-experienced professionals using AI to close the gap with senior colleagues. The organizations capturing this gain are investing aggressively in training the middle of their talent distribution.
Don't wait for the perfect implementation.
The WEF study found that companies with active AI programs, even imperfect ones, were already building institutional knowledge about what works in their specific context. Companies waiting for a cleaner strategy were falling behind in organizational learning — and organizational learning, not the tool, is the durable competitive advantage.
The Window Is the Point
Economic disruption on this scale doesn't arrive all at once. It arrives unevenly — sector by sector, function by function — moving fastest where task predictability and training data are richest. The 36-month window ahead is not the full arc of the disruption. It is when the firms and individuals who positioned correctly begin pulling away from the ones who didn't.
The retail parallel is instructive not because the net outcome was bad — it wasn't. It's instructive because the people who saw the internet as a channel to extend their existing model got the direction right and the implication catastrophically wrong. The ones who understood they were watching a change in the underlying economics — Amazon in 1996, Shopify in 2006 — built for the world as it was becoming.
You know which side of that you want to be on. The question is whether you've started acting like it.
